Technology Road Mapping of Two Machine Learning Methods for Triaging Emergency Department Patients in Australia

Abstract

Triaging is the categorization of patients attending the emergency departments (ED) into categories based on the patient's condition at arrival. In Australia, triage is a manual process, not guaranteed consistent and potentially error-prone. The problem with the manual process is that assigning incorrect triage categories to patients can result in delay of treatment for some patients. With the establishment of the Australian national health record system (MyHealth) and clinical data sharing standards such as HL7, it is possible to use patient history information as well as data about the patients' conditions at arrival in ED to quickly and accurately assign a triage category. The wide availability and application of machine learning (ML) methods, including medical applications using such methods, make these methods a possible solution to this problem. Before implementation of ML algorithms in triage, it is essential to understand the multiple dimensions of potential outcomes of health-services, including changes of clinical behaviors and workflows, social-economical-technical, and ethical and legal debates. This research uses Context-Content-Process (CCP), SWOT and Khoja--Durrani--Scott (KDS) frameworks to provide an initial review of a technical roadmap of the classification of patients attending emergency departments (ED) using different stages of Naive Bayes (NB) and Neural Network (NN) machine learning (ML) methods. This is the first research looking at the potential to use ML methods to assist in triage of patients in the Australia context considering outcomes. This research could be used to evaluate automation of the triage process or to support the manual process. The research results suggest that it is necessary to understand these multiple outcomes before future implementations are actually conducted.